How is the CPU score calculated?

[discstuz]

Just looking at the CPU score, here (please see the image).

Total A is 24163

Total B is 13404

B has a higher score for: Integer math, compression, CPU single thread, floating point math, SSE
A has a higher score for: Prime numbers, physics, encryption, sorting

I think only prime numbers and physics are significantly different.

Why are the totals so far apart?

[David (PassMark)]

Prime numbers result was 4x higher.
Physics was 5x higher.

The formula can be found here,
https://www.passmark.com/forum/perfo...-and-disk-mark

[Ryrynz]

Hi David,

How much do the SSE scores come into play? Intel always seems to have much lower scores in this area (3 vs 84 for the below processors)
Two machines I have here in store, the new quad core Pentium N5000 and the AMD dual core A6 9220, both have similar total scores 2304 vs 2542, yet the Pentium is far faster.. for example a winrar extraction is almost twice as fast, with twice the cores you expect this.. I would recommend re-evaluting how you calculate the total scores because I use your benchmark a lot in store and things just don't tend to stack up, the N5000 should be much higher..

[David (PassMark)]

How much do the SSE scores come into play?

It is a weighted harmonic mean. The V9 CPUMark formula is here.

Intel always seems to have much lower scores in this [SSE] area

It depends on what two CPUs you compare.
For example the AMD FX-8370 is a long long way behind the Intel i7-6700K.
But the i7-6700K is a long long way behind the Threadripper 1950X

The Intel Pentium Silver N5000 sucks so bad in the SSE test as it seems to be missing a bunch of critical AVX instructions. No AVX mean no vector acceleration. So in this respect the Pentium Silver N5000 is massively slower, and the benchmark correctly reflects this. In other areas, like encryption, sorting, physics, the N5000 is a quicker. The A6 9220 also seems to be better a single threaded tasks.

The SSE test result is a massive 20x difference. AVX brings a lot of performance to the table for tasks that can use it.

[Ryrynz]

Thanks, the CPu score though doesn't reflect general real world results, shouldn't the integer performance which is likely more reflective of real world Windows application performance be factored in a bit more heavily? The N5000 is on another level just being a modern quad core part.

[David (PassMark)]

CPu score though doesn't reflect general real world results

There is really no such thing as a single number which is "real world". There are 10,000s of different PC applications written in different languages, doing different things, compiled to different target architectures. Real world to one person might mean running Photoshop, but might mean running a web server to another user. So the comment is a bit meaningless.

At the moment there aren't a huge number of modern applications use AVX/FMA instructions. Some examples of those that do are, Video codecs, OpenSSL encryption (for HTTPS web pages), 3D rendering software like Blender, Scientific applications. But as more and more CPUs support it, it will become more common in software over the next few years.